Leveraging best-in-class AI technologies, INTNT.AI has boosted chatbot customer
satisfaction by three times for financial services companies.
- Chatbots have a notorious reputation for frustrating customers.
- A*STAR spinoff INTNT.AI uses a special breed of AI to significantly enhance the
performance of AI chatbots, and help them better understand what human users ask
- At the heart of INTNT.AI’s suite of tools are intent classification and emotion analysis
technologies developed in A*STAR.
Financial services companies are starting to pay attention to the performance of the chatbot
function in their customer-facing websites.
Chatbots are available 24/7 and can easily process questions from customers. The number of
customers that can be serviced at any one time is also not limited to the availability of human
customer service agents. This helps companies to save costs, especially if the chatbot relieves some
duties of the customer service and call centres.
Chatbots can also improve customer satisfaction and even boost sales, when they address questions
immediately and help the customers discover new products.
But most don’t. Typical AI chatbots have a 60-70% error rate from open questions. As a result, many
companies scale down or even shutter their bot programmes.
To meet the industry demand, A*STAR spinoff INTNT.AI has been turning around industry-first AI-
powered chatbots for insurance providers and banks in Singapore, with noticeable results.
INTNT.AI’s clients have seen their chatbot customer satisfaction go up by as much as 300%, chatbot
fails down by 85%, and user requests to speak to live agents from the chatbot down by 72%
In one client’s case, customers did a survey on their experience with their chatbot before and after
INTNT.AI’s improvements. The customer satisfaction scores went from 1.2 to 4.5, with 5 being the
What is INTNT.AI’s secret for its success?
INTNT.AI dramatically increases bot accuracy
TRAINING CHATBOTS TO UNDERSTAND INTENT
Mishandled intents, insufficient intents, and inefficient training are key reasons why many chatbots show poor performance and frustrate customers
Intent refers to the motive of the chatbot user, or in other words, what the user hopes to achieve by interacting with the chatbot. Since chatbots do not understand human language, they need to be “trained” to extract intent from the user's input and respond accordingly. Therefore, for a chatbot to be successful, it needs to be able to correctly identify the user's intent and offer a relevant, tailored response.
Improving a chatbot is a constant, ongoing process. This includes fixing false positives and false negatives, as well as identifying new intents.
INTNT.AI implemented an intent classification technology powered by A*STAR's Institute for Infocomm Research (I2R). The technology applies unsupervised clustering to group large amounts of unannotated utterances at a granular level, to discover and facilitate the creation of new intents.
Additionally, INTNT.AI pre-trained a hybrid machine learning and rules-based model using the latest natural language processing technology, which enabled INTNT.AI to process human language data and auto-extract meaning from it.
Therefore, chatbots enhanced by INTNT.AI’s training are equipped with a much more powerful AI functionality. This allows them to understand the unstated intent behind the vast permutations of possible user inputs. For example, chatbots using INTNT.AI’s tech quickly learn that words such as “buy” or “get” are often associated with the intent to purchase.
DESIGNING BETTER USER JOURNEYS
At the heart of INTNT.AI’s suite of tools is CrystalFeel, an emotion analysis tool developed by A*STAR’s Institute of High Performance Computing (IHPC). The software analyses how the customer communicates, to quantify their emotions, mood, and attitude.
It consists of five algorithms that analyse the emotional intensity expressed in a given text input, based on five psychologically meaningful dimensions: anger intensity, fear intensity, sadness intensity, joy intensity, emotional valence intensity. The data is then used to create strategies that will improve outcomes of the interaction.
INTNT.AI also utilised A*STAR's award-winning Speech-to-Text and Digital Emotion. This tool accurately transcribes what customers say in real-time, while taking into account local accents such as Singlish.
By understanding not only what the user’s intent is, but also the emotional intensity underlying that intent, INTNT.AI chatbots can give better and more accurate responses to users.
"Chatbots have developed a notorious reputation. Many of them cause frustration for both the company and its customers," remarked Manuel Ho, Founder and CEO of INTNT.AI. "We work with companies to turnaround their chatbot performance and make real business impact.”
A*STAR SUPPORTS START-UPS ON THEIR INNOVATION JOURNEY
Through programmes and platforms that cover the spectrum from strategy and networking to R&D partnerships and transformative technologies, A*STAR supports start-ups on their innovation journey and helps them to level up their business.
"For start-ups, innovation is key to gaining competitive advantage and making an impact in their respective sectors," explained Lee Han Boon, Director, A*STAR Enterprise. "A*STAR's technology injection enables start-ups like INTNT.AI to increase its technological differentiation and accelerate its go-to-market strategy."